`Sales Analysis, and ANNs for the Mass
`Appraisal of Residential Properties in
`Northern Ireland
`
`William J. McCluskey and Richard A. Borst
`
`The
`
`application of mass appraisal techniques
`for property tax assessment has become
`widespread throughout the world. For resi-
`dential properties, such techniques tend to
`rely on a variety of statistically based multivariate
`models. This paper applies three techniques of mass
`appraisal, namely, multiple regression analysis (MRA),
`comparable sales analysis, and artificial neural net-
`works (ANNs), to a data set of residential sales from
`the suburbs of Londonderxy, Northern Ireland. The
`objective is to analyze the performance of the models
`in terms of such criteria as predictive ability, explain-
`ability, and defensibility. The results of this research
`demonstrate the consistency of all three models in
`terms of predictive accuracy, while highlighting the
`variation among the models in terms ofease of use
`and taxpayer understanding.
`
`The Residential Property Tax in
`Northern Ireland
`The present system of property taxation in Northern
`Ireland involves the assessment of both commercial
`
`William J. McCluskey is a senior lecturer in real estate at
`the University cf Ulster, Northern Ireland, and Richard
`A.Borst is senior vice president, Cole-Layer- Trumble
`Company, West Chester, Pennsylvania.
`
`The statements made or views expressed by authors in
`Assessment Journal do not necessarily represent a policy
`position ofthe InternationalAssociation ofAssessing
`Officers.
`
`and residential property. The current basis of assess-
`ment is the net annual value (NAV), which can best
`be described as an open market rental value, as of a
`predesignated date, on the assumption that the tenant
`is responsible for all repairs and expenses. Tradition-
`ally, most properties were occupied under leases, and
`there was ample rental evidence on which to deter-
`mine NAys. For commercial property, occupancy is
`stilt primarily leasehold, so there currently are suffi-
`cient market rents to establish the rating assessment
`objectively, and, indeed, the 1997 general revaluation
`of commercial property will be based on this ap-
`proach.
`Residential property, on the other hand, is pre-
`dominately owner occupied. Therefore) establishing
`NAys based on a weak and almost nonexistent rental
`market could prove to be complexif not impossible.
`One would expect that any future revaluation of resi-
`dential property should be based on open market sell-
`ing prices. Because commercial property is being re-
`valued for the first time since 1976, there are no
`immediate plans to revalue the residential sector.
`However, some form of residential revaluation would
`be expected in order to ensure parity with the com-
`mercial sector. The alternative options would appear
`to be
`
`Abolish the residential property tax. However, the
`revenue loss would need to be replaced by an alter-
`native revenue source. None of the alternatives, in-
`cluding a poll tax, would have any real advantage
`over the present property tax.
`
`January/February 1997
`
`47
`
`
`
`"Factor-up" the existing residential NAVs to a level
`broadly equivalent to the increase in the
`cial sector. This broad-brush approach has the ad-
`vantage of simplicity, but has the disadvantage of
`exacerbating the value anomalies inherent within
`the present NAVs.
`Revalue all residential property to a NAy. As stated
`earlier, this would prove extremely difficult because
`of the relative scarcity of open market rental trans-
`actions.
`Undertake a revaluation on the basis of open marker
`capital values.
`
`appraiser to review large numbers of properties for
`judgmental factors such as condition, quality of con-
`struction, and ultimately a computer-generated esti-
`mate of value.
`
`Data
`The number of residential properties in Northern Ire-
`land is very large, as indicated in table I. With in ex-
`cess of 600,000 properties, how to value them needs
`to be considered within the context of the resources to
`undertake the task, the available time, and the reliabil-
`ity and defensibility of the results.
`
`The most realistic option would be to maintain a
`residential property tax, but based on capital values.
`This approach has a number of distinct advantages,
`including sufficient market evidence, enhanced tax-
`payer understanding of the system, and the ability to
`apply mass appraisal technologies.
`
`Mass Appraisal
`Because the property tax is an ad valorembased tax,
`the assessments must be accurate and must be updated
`periodically in order to meet the dual requirements of
`equity and fairness. It could be argued that the most
`appropriate appraisal approach for compliance with
`these requirements would be to consider each prop-
`erty individually and determine a discrete value, that
`is, a manual approach. However, because of the lack
`of statistics on the accuracy of manually derived as-
`sessments, it could be equally argued that mass ap-
`praisal is the only discipline that holds itself out to the
`rigors of statistical verification. Mass appraisal is a
`process by which a universe
`f properties is appraised
`using standardized techniques. The International As-
`sociation ofAssessing Officers (1990) has defined
`mass appraisal as die systematic appraisal of groups of
`homogeneous properties as of a given date using stan-
`dardized procedures and statistical testing. Early writ-
`ers, such as Renshaw (1958), Pendelton (1965),
`Eisenlauer (1968), and Stenehjem (1974), argued that
`mass appraisal, as opposed to single-property ap-
`praisal, requires the development of robust models ca-
`pable of replicating the components of value. Mass
`appraisal is distinguished from single-property ap-
`praisal by lending itself to statistical validation of the
`results and by usually being performed by teams of
`people with specialized skills. For example, data col-
`lection need not be performed by an appraiser.
`Rather, persons trained in data collection can cost-ef-
`fectively gather necessary descriptive information
`about a property, including its measurements, room
`counts, and architectural style. This allows the skilled
`
`48 Assessmentjourncal
`
`Table I
`Residential Property Figures: 1991-95
`
`Year
`
`1991
`1992
`1993
`1994
`1995
`
`Residential properties
`
`583,628
`589,937
`597,676
`606,753
`607,223
`
`$ Source: Department of Evaluation for Northern
`Ireland; Raring Division, Statistics
`
`Data for this research were supplied by the Valua-
`tion & Lands Agency. This government agency is re-
`sponsible for the assessment of all real property in
`Northern Ireland for property tax purposes. The data
`comprised all open market sales for residential prop-
`erty from October 1993 to September 1995 for the
`suburbs of Londonderry. The variables captured are
`described in table 2.
`
`Table 2
`Description of Variables
`
`Variables
`
`Description
`
`Selling price
`Transaction daec
`
`Floor area
`Beds
`Age
`
`Type
`Class
`Heating
`Garage
`Ward
`
`Actual price in £
`Converted to a reverse date of sale (P1)05)
`¡ea days from January 1, 199G
`Gross external area in square meters
`Number of bedrooms
`Date built expressed in five age categories:
`I is oldest, 5 is newest
`House, bungalow, chalet, terrace
`Semi-detached, detached, terrace
`Full, part, none
`Single, double, none
`Political boundary
`
`Several Factors considered critical to developing ac-
`curate estimates of capital value were not available for
`the analysis. Most notable are the absence of land. size,
`
`
`
`the quality of construction, the condition of the prop-
`erty, and indicators of remodeling or effective age.
`After obvious outliers were eliminated, the data set
`was comprised of 1,495 sales. The continuous vari-
`ables have the statistics described in table 3.
`After thirty wards were combined into five ward
`groups based on similar age and selling price per
`square meter, the categorical variables had the distri-
`butions shown in table 4.
`
`Table 3
`Statistks on Continuous Variables
`
`Price
`(5)
`
`44,277
`40,000
`15,256
`25,250
`123,000
`1,495
`
`Area
`
`Beds
`
`RDOS Age
`
`113
`104
`35
`51
`334
`1,495
`
`3.3
`3.0
`0.8
`0.0
`8.0
`1,495
`
`486
`509
`212
`94
`822
`1,495
`
`3.5
`4.0
`1.3
`1.0
`5.0
`1,495
`
`Mean
`Median
`Standard deviation
`Minimum
`Maximum
`Number
`
`Table 4
`Categorical Variables
`
`Description
`
`Code
`
`Number
`
`Single-scoiy
`Chalet
`Cottage
`Two-story
`
`Detached
`Semi-detached
`Terrace
`
`Full central heating
`Part central heating
`No central heating
`
`Double garage
`Single garage
`Outbuilding
`No garage
`
`Group I
`Group 2
`Group 3
`Group 4
`Group 5
`
`Buil&ng type
`
`RU
`CH
`CO
`HO
`
`Building class
`
`DEI'
`SDT
`TER
`
`Heating
`
`FCH
`PCE
`NCH
`
`Garage type
`
`MHD
`MHS
`OTB
`ASS
`
`Ward groups
`
`WI
`W2
`W3
`W4
`W5
`
`449
`185
`19
`842
`
`454
`608
`433
`
`895
`256
`344
`
`20
`483
`342
`650
`
`172
`46
`554
`83
`640
`
`Even after obvious outliers were eliminated, the
`data set had some obvious inconsistencies that limit
`the statistical accuracy of any value estimates devel-
`oped by any of the techniques subsequently applied
`and described here. Consider table 5. lt contains a
`small subset of properties that are identical in descrip-
`tion except for sale date, but that have widely different
`selling prices. This condition can be found extensively
`in the data set. The implication is that the statistical
`results will be limited by this inherent variability. It is
`believed that knowledge of land size, construction
`quality, condition, and effective age would make it
`possible to discriminate among these ptoperties, but
`this limitation of the data is considered a constraint of
`the analysis that follows. With the data presented in
`the table, the minimum coefficient of variation
`(COV) for Actual-Predicted that could be obtained by
`using the mean selling price as the estimate for ail
`properties in table 5 is 14.3 percent. It is possible that
`by adjusting for date of sale, the variability would be
`reduced. The Ad/price in the table was derived by ad-
`justing the selling price for date of sale using the coef-
`ficient for RDOS from a MBA formulation to be pre-
`sented subsequently. Even after adjusting for date of
`sales, the COY of Actual-Predicted using the mean of
`the A4price as the predicted is 11.5 percent. This re-
`suit is an indicator of the inherent variability within
`the data set.
`
`Analysis
`It is the intention of this paper to consider the mass
`appraisal option in terms of the application of three
`techniques:
`
`MRA
`comparable sales analysis
`ANN5
`
`Each technique will then be evaluated in terms of pre-
`dictive ability, explainabiliry, stability, repeatability,
`and defensibility.
`
`Ward Groups
`The initial data set contained thirty distinct wards. To
`facilitare the use of the several multivariate analytic
`techniques, it was appropriate to group the thirty
`wards into a smaller number of ward groups having
`similar characteristics. Table 6 summarizes the results
`of the effort, achieved by studying the patterns of
`price, size, and age and assigning each ward to one of
`five groups. The goal was to have similarity within a
`group and dissimilarity among groups. All analysis
`presented tises this ward group schema.
`
`Januory/Februory 1997 49
`
`
`
`Testing and Training Groups
`The original data set of 1,495 properties was parti-
`tioned into two major subsets of 1,346 and 149 prop-
`erties. The larger subset was used for the calibration of
`the MRA and ANN models as well as providing the
`"Sales" for comparable sales analysis. The smaller
`subset was used as a set-aside sample for evaluating
`the predictive ability of the various models. lt also be-
`came the "Subjects" data file for the comparable sales
`
`analysis procedure. The 1,495 properties were ar-
`ranged in a random order and every tenth property
`was placed in the smaller subset.
`
`MRA
`One of the most significant advances in terms of mass
`appraisal has been the development of MRA as a tool
`for the prediction of value (see Gloudemans and
`Miller 1978; Smeltzer 1985; and Fraser and Blackwell
`
`Table S
`Price Inconsistencies within the Data Set
`
`Price
`
`Adjprice
`
`Ward
`
`Area
`
`Class Age Type Beds Heating Garage RDOS
`
`36,000
`38,000
`38,000
`38,500
`41,000
`46,500
`52,500
`
`39,494
`47,782
`46,138
`48,638
`47,446
`49,034
`58,254
`
`Crevagh
`Crevagh
`Crevagh
`Crevagh
`Crevagh
`Crevagh
`Crevagh
`
`93
`93
`93
`93
`93
`93
`93
`
`SDT
`SOT
`SOT
`SOT
`SOT
`SDT
`SOT
`
`4
`4
`4
`4
`4
`4
`4
`
`HO
`HO
`HO
`1-10
`HO
`HO
`HO
`
`3
`3
`3
`3
`3
`3
`3
`
`FCH
`FCH
`l'CH
`FCH
`l'CH
`l'CH
`FCH.
`
`ABS
`ABS
`ABS
`ABS
`ABS
`ABS
`ABS
`
`255
`714
`594
`740
`472
`185
`420
`
`Table 6
`Analysis by Ward Groupings
`
`Ward
`group
`
`Avenge
`price
`
`Average
`price/area
`
`Average
`area
`
`Avenge
`age
`
`WI
`WI
`WI
`W'
`W2
`W2
`W2
`W3
`W3
`W3
`W3
`W3
`W3
`W3
`W3
`W3
`'073
`W3
`W3
`W4
`W4
`W5
`W5
`W5
`W5
`W5
`W5
`W5
`w5
`W5
`
`31,197
`34,161
`36,839
`37,188
`38,461
`38,798
`53,900
`31,519
`32,333
`36,235
`38,265
`38,277
`41,996
`43,317
`44,850
`46,914
`47,038
`48,447
`50,244
`58,046
`63,247
`35,796
`43,792
`44,827
`45,008
`47, 144
`48,083
`49,029
`50,016
`68 .64 1
`
`303.6
`318.7
`336.9
`342.2
`370.6
`370.2
`359.8
`321.7
`347.7
`382.9
`372.8
`394.4
`338.6
`401.7
`378.4
`383.4
`377.8
`361.4
`383.!
`445.4
`4 57.6
`418,8
`427.9
`414.5
`402.8
`433.3
`430.2
`512.2
`404.7
`404.1
`
`103.5
`109.7
`113.3
`111.6
`107.3
`107.i
`155.0
`97.9
`93.0
`94.9
`102.8
`97.5
`123.1
`108.4
`121.0
`124.0
`124.7
`135.0
`130.0
`132.4
`143.4
`87.2
`103.3
`108.9
`113.0
`109.8
`112.6
`97.4
`125.7
`169.3
`
`1.7
`1.1
`1.7
`1.4
`1.9
`2.2
`2.8
`4.0
`3.3
`3.3
`4.2
`4.2
`4.1
`4.3
`1.3
`3.8
`4.2
`5.0
`4.0
`3.7
`4.3
`4.0
`3.9
`4.4
`3.9
`4.0
`3.4
`3.8
`4.0
`2.8
`
`Ward
`flame
`
`Victoria
`The Diamond
`West la od
`Rosem ou n r
`Ebringron
`B eech wo od
`Bn agh e r
`Creggan Central
`Ceggan South
`Brandywell
`Caro Hill
`Cisnagelvin
`Claudy
`Aitnagelvin
`Strand
`H oli ym o u n t
`New buildings
`Shagi tallow West
`Eglin ton
`Caw
`Ballyn ash allog
`Kilfen na n
`Foyle Springs
`Crevagh
`Clondermor
`Springtown
`Pennyburn
`Shancallow East
`C u Im o re
`Enagh
`
`50 Assessment Journal
`
`
`
`1989). MBA models can take several functional forms
`depending on the type of variables being used and
`their interrelationships. For the purposes of this paper,
`two model formulations were evaluated. The first was
`a linear additive mode! of the form
`
`Price = a0+ a1x1 + ax2 + ... a,,x,.
`
`(1)
`
`The second was a multiplicative model of the form
`
`Price = b0x1b1x2h2. Xkbk+Ixk+j . .
`
`. b,/'n,
`
`(2)
`
`where the x for i = 1 to le are continuous variables
`and x for i = k +1 to n are categorical variables. The
`continuous variables represent those factors that can
`be treated numerically and have a demonstrated be-
`havior, that is, a marginal change in eath independent
`variable is expected to produce a corresponding in-
`crease or decrease in the dependent variable. The cat-
`egorical (binary) variables are used to represent the
`factors such as building type in which the above de-
`scribed assumption is not deemed appropriate. Ifa
`categorical variable has p possible choices) p - I bi-
`nary variables were included in the models calibrated
`by MBA to avoid !inear dependency in the model
`structure. For example, the five ward groups are repre-
`sented by four variables: Wi, WZ W3, and W4. The
`linear model was calibrated twice. In the first calibra-
`tion, all possible variables were included. In the sec-
`ond, only certain variables were considered for the
`model calibration process. For example, Age was not
`used because it was found to be statistically insignifi-
`cant in predicting value. Beds was e!iminated for the
`same reason. The coefficients that were derived from
`the 1,346 sales in the training set are shown in table 7.
`
`The multiplicative model developed from the same
`sales data can be expressed as
`
`Price = 4,5 14(A rea 0.639 x RDOS 0.123
`x o.9o4TEEx i.i29S)Tx 0901ABS
`X o.9110T11X i.090 Wix l23Wj
`
`(3)
`
`Once again, in che multiplicative model only statisti-
`cally significant terms are included. lt turns out that
`although this model uses the least number of factors,
`it was the most accurate in terms of predictive ability.
`
`ANNs
`Recently, ANN models have been applied with vary-
`ing degrees of success to real estate problems (see Do
`and Grudnitski 1992; Tay and Ho 1994; Borst 1995;
`
`Table 7
`Coefficients for MRA Models
`
`Name
`
`Type
`
`MMI
`Coefficient
`
`MRA2
`Coefficient
`
`Intercept
`Area
`Age
`Beds
`RDOS
`TER
`SDT
`NCH
`FCH
`ABS
`OTB
`MHS
`HO
`CO
`BU
`WI
`W2
`W3
`W4
`
`Constant
`Continuous
`Conünuous
`Continuous
`Coniinuous
`Binary
`Bina7
`Binary
`Binary
`Binary
`Binary
`Binary
`Binary
`Binary
`Binary
`Binary
`Binary
`Binary
`Binary
`
`24,667
`303
`122
`861
`13.7
`4,474
`5,404
`1,792
`2,196
`11,693
`11,718
`8,924
`5,212
`4,418
`6,737
`3,978
`379
`1,358
`7,472
`
`26,507
`288
`N/A
`NIA
`13.7
`4,311
`5,699
`N/A
`N/A
`11,757
`12,052
`8,971
`N/A
`N/A
`1,810
`3,696
`N/A
`N/A
`7,176
`
`Evans, James, and Collins 1995; and Worzala, Lenlc,
`and Silva 1995). The same data sets used in MBA
`were processed via the ANN. In ANN terminology
`there were twenty-three input neurons, thirteen hid-
`den neurons, and one output neuron. There are five
`more candidate inputs for the ANN as compared to
`the MRA because the ANN has no assumption of in-
`dependence for the predictor variables; therefore, all
`categorical variables were included in the model for-
`mulation. In MBA, one each of the five categorical
`variables was eliminated to avoid linear dependence
`among the dependent variables. The solution to this
`ANN model formulation is expressed via a set of 32G
`coefficients that do not lend themselves to easy inter-
`pretation. Their presentation is omitted bçcause it
`would not be particularly illuminating. Statistical per-
`formance of the MBA and ANN model formulations
`are presented in a later section.
`
`Comparable Sales Analysis
`The most predominant method for the mass appraisal
`of residential properties is the sales comparison ap-
`proach: This is so primarily because the comparable
`sales analysis report developed for each property is
`easier to explain and defend to the general public than
`is almost any other method, for example, an equation
`calibrated by MRA.
`Each subject property is valued, using computçr-
`based techniques, by selecting several comparable
`properties that have recently sold. The selling prices
`are adjusted for differences in characteristics, location,
`
`January/February 1997
`
`51
`
`
`
`and sale date between each comparable and the sub-
`jecr. A final estimate of value is computed using the
`sevral (usually three to five) estimates obtained in this
`process. The method for selecting comparable saie
`properties, adjusting their selling prices, and comput-
`ing a final value estimate may vaiy among mass ap-
`praisai systems, but MBA is the most common
`method for determining the adjustments between a
`subject property and its comparable sates.
`The comparable sales analysis procedure may be
`viewed as the following four-part process:
`
`For a given subject property, find the n most com-
`parable sales.
`Adjust the selling prices of the comparable sales to
`match the characteristics of the subject.
`Use the several estimates of value to arrive at an
`estimate of market value.
`Present the results in a report format suitable for
`viewing or printing.
`
`The process of finding comparable sales uses dis-
`tance" to establish a measure of comparability between
`the subject and the comparable sale (Comp) under
`consideration. It is computed by weighting the differ-
`ences in chíracteristics brween the two.
`
`The distance, D, is calculated as follows:
`
`In this instance, the five sales with the lowest dis-
`tance are selected. Typically, X = 2, meaning that dis-
`tance takes on the form of a square root of the sum of
`the squares formulation. A sample set of parameters is
`presented in table 9.
`
`Table 9
`Variable Weightings
`
`Variable
`
`Weight
`
`Type
`
`Ward group
`Area
`Class
`Age
`Type
`Bedrooms
`Hearing
`Garage
`RDOS
`
`150
`15
`100
`50
`75
`50
`20
`20
`2
`
`Categorical
`Continuous
`Categorical
`Continuous
`Categorical
`Continuous
`Categorical
`Categorical
`Continuous
`
`The factor weights are chosen considering the
`magnitude of the variable itself. For example, ifa
`comparable sale is not in the same ward group, the
`contribution (before raising to the power of X) to the
`calculation is 150 X 1, or 150. If the size differs by
`ten square meters, the contribution is the same, that
`is, 15 X IO = 150. The weight for RDOSis 2 because
`the range and magnitude of the factor are quite large.
`For each comparable property, the sale price is ad-
`justed to the subject property as follows:
`
`D
`
`=
`
`[A(X1X,)f' +
`
`Xd),
`
`(4)
`
`Adjusted sale price = Sale price - (Comp MM
`- Subject MM),
`
`(5)
`
`where the terms are as explained in table 8.
`
`or alternatively for the ANN model:
`
`Adjusted sale price = Sa/e price - (Comp ANN
`- Subject ANN).
`
`(6)
`
`Given the several comparable sales, several adjusted
`selling prices are obtained. A weighted estimate
`(WghtEst) is formed as follows:
`
`WghtEst=ASP1,
`1=JW
`
`(7)
`
`where weight for Comp i = Wj and
`
`J
`
`_\+D2+(2D
`(D2
`t2)
`
`ASP1SP1I'2
`
`p1
`
`)
`
`Table 8
`Explanation of Terms
`
`Term
`
`Explanation
`
`X
`
`A1
`
`X,
`S
`
`= Minkowski exponent lambda
`= weight associated with the kh continuous characteristic
`value of che ah characteristic in the sale property
`value of ah characteristic in subject property
`= summation oí terms of icharacteristics
`
`= weight associated with thejth categorical characteristic
`= value ofjrh characteristic in sale property
`value ofjth characteristic in subject property
`summation 0f terms off characteristics
`
`X1
`Xsj
`I
`
`J 3
`
`(a,b) = lnversc dclta function(O, if a=6 1, ifa4)
`
`52 Assessnient Journal
`
`
`
`where Ai is the predicted price for property i, Si is
`the actual selling price for property i, Ai/Si is the
`ndividual property ratio, (A/51 raed is the median ap-
`praisal ratio, and A/S is the average appraisal-to-sale
`ratio.
`
`As table il illustrates, the statistical comparison is
`very close and would not lead to a clearly superior
`choice. It is interesting that MBA2 with fewer terms
`actually performs slightly better than MRAI and the
`multiplicative model is the best in terms of statisticai
`predictive accuracy. It also demonstrates that the
`ANN model performed at least as well as the other
`models.
`
`Table 11
`Comparison of Model Results
`
`MItAI
`MRA2
`MULTMRA
`ANN
`MRACOMPS
`ANNCOMPS
`
`R2
`
`0.772
`0.767
`0.801
`N/A
`N/A
`N/A
`
`COY (%)
`
`COD (%)
`
`14.0
`13.7
`13.1
`14.0
`14.4
`14.2
`
`11.2
`11.0
`10.5
`11.1
`11.0
`'l.i
`
`McCluskey 1996). The technique lacks transparency
`in its failure to explain adequately how the results
`were obtained. In application terms, it is easier to ap-
`ply than regression because linearity is not assumed,
`allowing all variables to be used without the need for
`transformation. Repeatability, however, can be a sig-
`nificant problem, in that two different ANN packages
`can produce differing results; this is primarily due to
`the nature of the ANN in randomly setting the initial
`weightings.
`M a practical matter, the comparable sales analysis
`presentation format is most suitable for review by ap-
`praisers and property owners not familiar with statisti-
`cal methods. It offers the most explainable and defen-
`sible presentation of an individual appraisal estimate.
`In conclusion, all the models tested achieved broadly
`similar results in relation to predictive ability. Varia-
`tions among the techniques are more apparent when
`repeatibility and taxpayer understanding are con-
`cerned. Therefore, overall, based on this research, one
`would have to conclude that the comparable sales
`analysis approach provides for both a statistically
`based algorithm for assessment and an easily under-
`standable end product.
`
`Conclusions
`Although the data set is known to have limitations,
`the predictive accuracy is encouraging across all the
`models. Ifa choice must be made among the methods,
`the multiplicative MBA model offers the best accuracy
`for the given data set. Therefore, in terms of predictive
`ability there is little to choose between the models.
`This said, it is noted that the practice of mass ap-
`praisal relies on the review of computer-generated esti-
`mates of value by competent personnel. M such, the
`differences in predictive accuracy of the methods
`evaluated in this analysis are likely to be mitigated by
`the review process.
`MBA does offer an element of transparency in that
`the model develops individual coefficients that form
`the basis of the predictive ability. M a general rule,
`MBA models incorporating variables that have been
`subject to transformation often produce better predic-
`tive results. However, such models become more diffi-
`cult to defend in courts and tribunals, and taxpayer
`understanding is also somewhat reduced. One advan-
`tage that MBA models do have relates to repeatability,
`in that two different statistical packages using the
`same data will give the same results; therefore, one has
`confidence in the results.
`The predictive abilities of ANNs have been estab-
`lished through several investigative studies (Borst and
`
`-
`
`References
`Borst, R. A. 1995. Artificial neural networks in mass
`appraisal. Journal ofProperty Tax Assessment &Ad-
`ministration 1(2): 5-15.
`
`Borst, R. A., and W. J. McCluskey. 1996. In Flaherty,
`J., R. Lombardo, P. Morgan, and B. M. de Silva
`(eds.). Artificial neural networks in quantitative
`methods in property. Melbourne, Australiv Royal
`Melbourne Institute of Technology.
`
`Do, A.Q, and G. Grudnitski. (December) 1992. A
`neural network approach to residential property
`appraisal. The Real Estate Appraiser 58(3):38-45.
`
`Eisenlauer, J. F. (October) 1968. Mass versus indi-
`vidual appraisals. App raisal Journal 36(4):532-40.
`
`Evans, A., H. James, and A. Collins. 1995. ArtifIcial
`neural networks: An application to residential valu-
`ation in the UK. Journal ofProperty Tax Assessment
`&Administration 1(3) :78-92.
`
`Fraser, R. R., and F.M. Blackwell. 1988. Compa-
`rable selection and multiple regression in estimating
`real estate value: An empirical study. Journal of
`Valuation 7(3):1 84-201.
`
`54 Assessment Journol
`
`
`
`Gloudemans, R. J., and D. W. Miller. 1978. Multiple
`regression analysis applied to residential properties:
`a study of structural relationships over time) Deci-
`sion Sciences 7:294-304.
`
`International Association of Assessing Officers. 1990.
`Property appraisal and assessment administration.
`Chicago: International Association of Assessing
`Officers.
`
`Pendleton, W. (January) 1965. Statistical inference in
`appraisal and assessment procedures. App raisalJour.
`na! 33(1):73-82.
`
`Renshaw, E. F. (December) 1958. Scientific appraisal.
`National Tax Journal 11:314-22.
`
`Smeltzer, M. V. 1985. The application of multi-linear
`regression analysis and corrdation to the appraisal
`of real estate. Appraisal Review 28:1-13.
`
`Stenehjem, E. 1974. A scientific approach ro the mass
`appraisal of residential property. In Automated Mass
`App raisal ofReal Property. Chicago: International
`Association of Assessing Officers.
`
`Tay, D. P., and D. K. Ho. 1994. Intelligent mass ap-
`praisal. Journal ofProperty Tax Assessment ¿rAdmin-
`istration l(1):5-26.
`
`Worzala, E M Lenk, and A. Silva. 1995. An explora-
`tion of neural networks and its application to real es-
`tate valuation. Journal ofReal Estate Research
`10(2):135-201.
`
`PMC Has Your GIS ProjEct Solution
`Innovation, integrity, and customer service...
`our commitment to providing you with the best.
`the GIS industry has to offer.
`
`We provide
`Cost competitive, county-wide GIS projects
`Automated CSR and use based agricultural land assessment
`Customized parcel database querying
`Map based spatial analysis tools
`Qigital aerial orthophotography
`DGPS data collection and database linking
`AutoCAD R13 DWG, binary DXFT and MicroStation DON
`file import and integration
`Custom maps, and more
`PMC associates have more than 25 years experience in the creation of digital maps, map databases, and enterprise
`Geographic Information Systems projects. We'll work with you to develop solutions that meet your digital mapping and
`budget goals. Put your GIS projects on the leading edge with l'MC, ProMap Corporation.
`
`1*
`
`iwaS id
`
`ri1un
`
`pMc
`PTuMap Corporation
`1531 ftjtpo.t R Stt a
`Mie. Iowa s000
`
`For More Information
`Phone: SE.2333311
`E-mail: info@promap.com
`
`Jonuory/February 1997
`
`55